If Every Car Were a RAV4, 30,311 Fewer People Would Die Each Year

NHTSA announced in April that 36,640 people died on American roads in 2025, which the agency called encouraging because it represented a 6.7% decline from 2024 and pushed the fatality rate to 1.10 per 100 million vehicle miles traveled, the lowest since before COVID.^[1] Everyone celebrated. Nobody ran the next calculation.

The Toyota RAV4 kills its occupants at a rate of 0.19 per 100 million VMT. That is not a typo: over the ten-year FARS window from 2014 through 2023, the RAV4 accumulated 914 occupant deaths across 1,834 fatal crash involvements in a registered fleet of 3.76 million vehicles.^[2] The national average is 1.10. Put differently, its rate is 5.8 times lower than the national average.

So: Americans drove roughly 3.33 trillion miles in 2025. At the national average rate, 36,640 of them died. At the RAV4's rate, 6,329 would have. Put another way, 30,311 people are dead because the American vehicle fleet is not composed entirely of Toyota RAV4s, which is simultaneously the most absurd sentence in automotive journalism and the most precisely true one I have ever written.

This is a thought experiment, not a policy proposal. Obviously. But the number it produces is real, and the math behind it is trivially verifiable: divide 2025 deaths by the 2025 rate to recover VMT, then multiply by the RAV4's FARS rate, and a calculator will do the rest. The result says that 82.7% of all American traffic fatalities in 2025 occurred in vehicles whose per-mile death rate exceeds the rate already achieved by a mass-market crossover that starts at $32,350.

Make it more realistic. Keep the segments intact. In sedans, the safest mass-market option is the Kia Forte at 0.40 per 100 million VMT. Among pickups, the Ram 1500 leads at 0.13. For vans, the Dodge Caravan sits at 0.23. Weight these by approximate fleet share and the composite "best-in-class-everywhere" rate comes to 0.305 per 100 million VMT, roughly 3.6 times safer than the actual fleet, producing a hypothetical 10,159 deaths per year instead of 36,640. That is a reduction of 26,481 lives annually, achieved by nothing more radical than every buyer in every segment choosing the model that already exists and already has the lowest death rate within its class.

Now flip it. Run the same calculation with the worst-in-class models in each segment, vehicles like the Nissan Maxima at 5.11 per 100M VMT for sedans, the Chevrolet Tracker at 7.83 for SUVs, and the Chevrolet S-10 at 4.83 for pickups. A fleet composed of these produces a composite worst-case rate of 5.23 and a hypothetical 174,160 annual deaths, quadrupling reality. Between the best-in-class fleet and the worst-in-class fleet, the spread is a factor of 17, and both contain only real vehicles that real Americans drove during the FARS window.

Within a single vehicle class, the numbers are just as ugly. SUV fatality rates span from 0.19 (RAV4) to 7.83 (Tracker), a 41-to-1 ratio between models that occupy the same showroom category and, in some cases, the same price bracket. Pickups run from 0.13 (Ram 1500) to 4.83 (S-10), a 37-to-1 spread. Sedans stretch from 0.40 (Forte) to 5.11 (Maxima), which is 12.8-to-1. These are not experimental vehicles versus production models. They are all mainstream nameplates you can find in any used car lot in the country, and one will kill you at 41 times the rate of another.

Why this does not mean what you want it to mean

A vehicle's fatality rate is not a measurement of its engineering alone, but rather a composite of the vehicle's structure, its safety equipment, its weight, its typical driving environment, the demographics of its buyer, the behavior of its average driver, and the vehicles it tends to collide with. The RAV4's rate of 0.19 is partly Toyota's engineering and partly the fact that RAV4 buyers disproportionately live in suburbs, commute on well-maintained roads, and belong to demographic brackets that drink less, speed less, and crash less.

If you took the entire population of Nissan Maxima drivers, people skewing younger and more urban, with higher average speeds and higher impairment rates, and placed them all in RAV4s, the RAV4's rate would climb. How much? Nobody knows, because FARS does not decompose the contribution of vehicle engineering versus driver behavior to per-model rates. That decomposition is the single most important number in vehicle safety, and it does not exist in any public dataset.

There is a second objection, and it is stronger. In a fleet of identical RAV4s, the weight-class advantage that currently shields RAV4 occupants in collisions with lighter sedans would vanish. Every multi-vehicle crash would become a fair fight. Physics guarantees that the RAV4's rate in a homogeneous fleet would be higher than 0.19, because some of the 0.19 was purchased with the misery of Civic and Corolla occupants on the other side of the crash. At roughly 3,600 pounds, the RAV4 sits comfortably in the middle of the weight spectrum, heavy enough to survive collisions with Civics and Corollas, light enough to avoid the diminishing-returns zone above 4,000 pounds. A fleet where every vehicle weighs 3,600 pounds would eliminate the mass asymmetry that currently kills about 4,200 people per year in lighter vehicles hit by heavier ones.^[3] Some of those deaths would redistribute upward into the now-unprotected RAV4 occupants.

What the number actually means

Even after discounting for demographics, behavior, and weight-class effects, the residual gap between the best and worst vehicles within a class is enormous, and the engineering component of that gap is not zero. Consider that the Kia Forte and the Nissan Maxima sell to overlapping demographics in overlapping geographies, yet one has a death rate 12.8 times the other. Removing the 2005-era nameplates from the worst-case list barely moves the needle: the Chevrolet Impala (discontinued 2020, rate 5.0) and the Chevrolet Cobalt (discontinued 2010, rate 5.1) are recent enough that their safety deficits reflect engineering decisions made this century.

Vehicle choice is not the whole story. But 30,311 is the upper bound on what it costs, and even a quarter of that number would be the equivalent of an entire decade of AEB mandate savings delivered overnight. NHTSA projects AEB will save 360 lives per year when fully implemented by 2029.^[4] If vehicle choice contributes even 10% of 30,311, that is 3,031 lives: eight years of AEB benefits in a single variable that is already under consumer control, right now, at every dealership in the country.

What to do with this

Before signing a lease, check the FARS-derived fatality rate for the specific model you are considering, not the star rating, not the IIHS award, but the actual per-VMT death rate computed from a decade of real crash outcomes. A five-star NHTSA rating tells you how the vehicle performs in a controlled lab test. the FARS rate tells you how often its occupants actually die on real roads. Frequently, these two numbers disagree. If your target model has a rate above 2.0 and a competitor in the same class is below 0.5, the data says you are paying for a vehicle that is four times as likely to kill you per mile driven, and no amount of advanced driver-assistance features printed on the window sticker changes what the decade-long population-level outcome data shows. Check VINs for open recalls at nhtsa.gov/recalls, and compare IIHS ratings at iihs.org/ratings. But start with the fatality rate. It is the only number that counts corpses instead of crash-test dummies.

Limitations

The fleet homogenization calculation assumes uniform VMT distribution and does not account for multi-vehicle crash dynamics in a homogeneous fleet. FARS per-model rates use estimated VMT from NHTS fleet surveys, introducing approximately ±15% uncertainty for models with smaller registered fleets. NHTSA's 2025 national death figure of 36,640 is an early estimate, not a final count. This analysis cannot decompose the rate into vehicle-engineering versus driver-behavior components because no public dataset contains both vehicle identification and granular driver demographics at the individual crash level. Some of the "vehicle choice" effect captured here is actually a driver-selection effect, and the two are inseparable in available data.

Strongest counterargument

that thought experiment treats per-model fatality rates as properties of the vehicle when they are, in reality, properties of the vehicle-driver-environment system. Consider the Nissan Maxima's rate of 5.11: that number is not a statement about the Maxima's steel and airbags alone, but a statement about who buys Maximas, where they drive them, how fast, and how sober. A Maxima sold exclusively to 45-year-old suburban commuters would almost certainly have a lower rate. To whatever extent this demographic confound explains the inter-model spread, the 30,311 figure overstates the contribution of vehicle choice and understates the contribution of driver behavior, road infrastructure, and enforcement. Honestly, we do not know the split, and FARS was not designed to tell us.